In a recent article in ScienceDaily they discussed some interesting findings. The researchers performing the test had 21 infants (11 months old) stare at a screen with a butterfly on it. The screen would introduce additional visual distractions and as soon as the child’s eyes left the butterfly, it would disappear. Over the course of the 15 day (5 visit) experiment, the child would focus (or concentrate) longer and longer on the butterfly over time. They found that as the child was trained to concentrate better, they showed improvements in other tasks as well such as pattern recognition. The researchers did note that in adults, the same kind of cross-functional benefits are not seen when they improve their abilities in a specific task. The same kind of improvement in concentration and pattern recognition was not seen in the other 21 infants in the experiment (the control group) who were just watching TV for 5 visits.
For those of us who are AI enthusiasts and are looking to the brain for hints, this provides another nugget of insight. It shows that fundamental skills are strongly inter-related and your ability to learn a task/skill is not independent from learning other tasks/skills. Given that the same result is not also present in adults, it shows that there is an element of “building the base” in the brain early on in life and that base of core skills is used to learn more advanced skills later in life. It supports the “you can’t teach an old dog new tricks” philosophy in that when you are older you can’t change the base of knowledge you are working with to learn new skills. It’s as if you build the tools you’ll need as an infant and then you’re stuck with those for the rest of your life.
As part of a project named SyNAPSE, IBM announced that it has built prototype chips which mimic human brain function. It appears these are hardware neural network chips with some additional functionality. IBM has stated that these chips were designed based on human brain function with simulated neurons, axons, and synapses.
The chips are still early in development as they have not yet been used for any advanced functionality which a standard computer processor is not capable of. The chips were only used in test cases such as playing pong against a human competitor, navigation tasks, and pattern recognition.
The project appears to have good backing with 21 million in funding from DARPA and obviously good financial support from IBM, one of the largest technology companies.
Check out the project site for some interesting reading and some videos about the project:
Below is the 2011 AAAI video competition winner. It shows some very interesting and advanced capabilities and the creators of this video and robot swarm deserve a lot of credit for their accomplishment.
I will be covering more on swarm concepts in future posts, and I plan to dissect this one to understand more about it’s capabilities and where they are working on improving it.
Stanford recently announced that they are making several courses available online and free to the world. These aren’t just the usual pre-recorded YouTube style classes, but instead are run just like any other e-learning course, except in this case, to potentially hundreds of thousands of people. They say the courses will have scheduled classes, homework, feedback on your work, and even a certificate of achievement from Stanford University for completing the course.
The following two AI related courses are being offered in this online program:
Introduction to Artificial Intelligence
This course offers a general introduction to the field of Artificial Intelligence. It is put on by two pretty interesting teachers; Sebastian Thrun a professor from Stanford and Peter Norvig who is Director of Research at Google. A lot of the hype around this course is likely due to those who are teaching it. At the time of this writing, there are 135,643 people registered for more information on the course. It says that signup is “temporarily unavailable”, likely due to overwhelming demand. More information on this class can be found at http://www.ai-class.com/.
Introduction to Machine Learning
The concept of machine learning in this course has to do with machines that learn information or concepts which were not explicitly programmed in. At the time of this writing, there are 38,886 people signed up for additional information about the Machine Learning course. The course is taught by professor Andrew Ng and covers many of the common machine learning concepts from theory to practical implementation. The course description and a place to sign up for more details can be found at http://www.ml-class.com/.
Two very interesting classes that may interest AI enthusiasts all over the world. A great move by Stanford to try what they’ve called a “bold experiment in distributed education”. The fact that it focuses on AI and machine learning makes it even better.
